Commit | Line | Data |
---|---|---|
3fcfab16 AM |
1 | |
2 | #include <linux/wait.h> | |
3 | #include <linux/backing-dev.h> | |
03ba3782 JA |
4 | #include <linux/kthread.h> |
5 | #include <linux/freezer.h> | |
3fcfab16 | 6 | #include <linux/fs.h> |
26160158 | 7 | #include <linux/pagemap.h> |
03ba3782 | 8 | #include <linux/mm.h> |
3fcfab16 AM |
9 | #include <linux/sched.h> |
10 | #include <linux/module.h> | |
cf0ca9fe PZ |
11 | #include <linux/writeback.h> |
12 | #include <linux/device.h> | |
455b2864 | 13 | #include <trace/events/writeback.h> |
cf0ca9fe | 14 | |
c3c53206 JA |
15 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); |
16 | ||
26160158 | 17 | struct backing_dev_info default_backing_dev_info = { |
d993831f | 18 | .name = "default", |
26160158 | 19 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
26160158 JA |
20 | }; |
21 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
cf0ca9fe | 22 | |
5129a469 JE |
23 | struct backing_dev_info noop_backing_dev_info = { |
24 | .name = "noop", | |
976e48f8 | 25 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
5129a469 | 26 | }; |
5129a469 | 27 | |
cf0ca9fe | 28 | static struct class *bdi_class; |
cfc4ba53 JA |
29 | |
30 | /* | |
181387da | 31 | * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side |
cfc4ba53 JA |
32 | * locking. |
33 | */ | |
03ba3782 | 34 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 35 | LIST_HEAD(bdi_list); |
03ba3782 | 36 | |
839a8e86 TH |
37 | /* bdi_wq serves all asynchronous writeback tasks */ |
38 | struct workqueue_struct *bdi_wq; | |
39 | ||
018a17bd | 40 | static void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2) |
f758eeab CH |
41 | { |
42 | if (wb1 < wb2) { | |
43 | spin_lock(&wb1->list_lock); | |
44 | spin_lock_nested(&wb2->list_lock, 1); | |
45 | } else { | |
46 | spin_lock(&wb2->list_lock); | |
47 | spin_lock_nested(&wb1->list_lock, 1); | |
48 | } | |
49 | } | |
50 | ||
76f1418b MS |
51 | #ifdef CONFIG_DEBUG_FS |
52 | #include <linux/debugfs.h> | |
53 | #include <linux/seq_file.h> | |
54 | ||
55 | static struct dentry *bdi_debug_root; | |
56 | ||
57 | static void bdi_debug_init(void) | |
58 | { | |
59 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
60 | } | |
61 | ||
62 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
63 | { | |
64 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 65 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
66 | unsigned long background_thresh; |
67 | unsigned long dirty_thresh; | |
68 | unsigned long bdi_thresh; | |
345227d7 | 69 | unsigned long nr_dirty, nr_io, nr_more_io; |
f09b00d3 JA |
70 | struct inode *inode; |
71 | ||
345227d7 | 72 | nr_dirty = nr_io = nr_more_io = 0; |
f758eeab | 73 | spin_lock(&wb->list_lock); |
7ccf19a8 | 74 | list_for_each_entry(inode, &wb->b_dirty, i_wb_list) |
c1955ce3 | 75 | nr_dirty++; |
7ccf19a8 | 76 | list_for_each_entry(inode, &wb->b_io, i_wb_list) |
c1955ce3 | 77 | nr_io++; |
7ccf19a8 | 78 | list_for_each_entry(inode, &wb->b_more_io, i_wb_list) |
c1955ce3 | 79 | nr_more_io++; |
f758eeab | 80 | spin_unlock(&wb->list_lock); |
76f1418b | 81 | |
16c4042f WF |
82 | global_dirty_limits(&background_thresh, &dirty_thresh); |
83 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); | |
76f1418b MS |
84 | |
85 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
86 | seq_printf(m, | |
00821b00 WF |
87 | "BdiWriteback: %10lu kB\n" |
88 | "BdiReclaimable: %10lu kB\n" | |
89 | "BdiDirtyThresh: %10lu kB\n" | |
90 | "DirtyThresh: %10lu kB\n" | |
91 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 92 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
93 | "BdiWritten: %10lu kB\n" |
94 | "BdiWriteBandwidth: %10lu kBps\n" | |
95 | "b_dirty: %10lu\n" | |
96 | "b_io: %10lu\n" | |
97 | "b_more_io: %10lu\n" | |
98 | "bdi_list: %10u\n" | |
99 | "state: %10lx\n", | |
76f1418b MS |
100 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
101 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f7d2b1ec JK |
102 | K(bdi_thresh), |
103 | K(dirty_thresh), | |
104 | K(background_thresh), | |
c8e28ce0 | 105 | (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)), |
f7d2b1ec | 106 | (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)), |
00821b00 | 107 | (unsigned long) K(bdi->write_bandwidth), |
f7d2b1ec JK |
108 | nr_dirty, |
109 | nr_io, | |
110 | nr_more_io, | |
c1955ce3 | 111 | !list_empty(&bdi->bdi_list), bdi->state); |
76f1418b MS |
112 | #undef K |
113 | ||
114 | return 0; | |
115 | } | |
116 | ||
117 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
118 | { | |
119 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
120 | } | |
121 | ||
122 | static const struct file_operations bdi_debug_stats_fops = { | |
123 | .open = bdi_debug_stats_open, | |
124 | .read = seq_read, | |
125 | .llseek = seq_lseek, | |
126 | .release = single_release, | |
127 | }; | |
128 | ||
129 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
130 | { | |
131 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
132 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
133 | bdi, &bdi_debug_stats_fops); | |
134 | } | |
135 | ||
136 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
137 | { | |
138 | debugfs_remove(bdi->debug_stats); | |
139 | debugfs_remove(bdi->debug_dir); | |
140 | } | |
141 | #else | |
142 | static inline void bdi_debug_init(void) | |
143 | { | |
144 | } | |
145 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
146 | const char *name) | |
147 | { | |
148 | } | |
149 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
150 | { | |
151 | } | |
152 | #endif | |
153 | ||
cf0ca9fe PZ |
154 | static ssize_t read_ahead_kb_store(struct device *dev, |
155 | struct device_attribute *attr, | |
156 | const char *buf, size_t count) | |
157 | { | |
158 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 159 | unsigned long read_ahead_kb; |
7034ed13 | 160 | ssize_t ret; |
cf0ca9fe | 161 | |
7034ed13 NJ |
162 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
163 | if (ret < 0) | |
164 | return ret; | |
165 | ||
166 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
167 | ||
168 | return count; | |
cf0ca9fe PZ |
169 | } |
170 | ||
171 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
172 | ||
173 | #define BDI_SHOW(name, expr) \ | |
174 | static ssize_t name##_show(struct device *dev, \ | |
175 | struct device_attribute *attr, char *page) \ | |
176 | { \ | |
177 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
178 | \ | |
179 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
d9e1241e GKH |
180 | } \ |
181 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
182 | |
183 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
184 | ||
189d3c4a PZ |
185 | static ssize_t min_ratio_store(struct device *dev, |
186 | struct device_attribute *attr, const char *buf, size_t count) | |
187 | { | |
188 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 189 | unsigned int ratio; |
7034ed13 NJ |
190 | ssize_t ret; |
191 | ||
192 | ret = kstrtouint(buf, 10, &ratio); | |
193 | if (ret < 0) | |
194 | return ret; | |
195 | ||
196 | ret = bdi_set_min_ratio(bdi, ratio); | |
197 | if (!ret) | |
198 | ret = count; | |
189d3c4a | 199 | |
189d3c4a PZ |
200 | return ret; |
201 | } | |
202 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
203 | ||
a42dde04 PZ |
204 | static ssize_t max_ratio_store(struct device *dev, |
205 | struct device_attribute *attr, const char *buf, size_t count) | |
206 | { | |
207 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 208 | unsigned int ratio; |
7034ed13 NJ |
209 | ssize_t ret; |
210 | ||
211 | ret = kstrtouint(buf, 10, &ratio); | |
212 | if (ret < 0) | |
213 | return ret; | |
214 | ||
215 | ret = bdi_set_max_ratio(bdi, ratio); | |
216 | if (!ret) | |
217 | ret = count; | |
a42dde04 | 218 | |
a42dde04 PZ |
219 | return ret; |
220 | } | |
221 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
222 | ||
7d311cda DW |
223 | static ssize_t stable_pages_required_show(struct device *dev, |
224 | struct device_attribute *attr, | |
225 | char *page) | |
226 | { | |
227 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
228 | ||
229 | return snprintf(page, PAGE_SIZE-1, "%d\n", | |
230 | bdi_cap_stable_pages_required(bdi) ? 1 : 0); | |
231 | } | |
d9e1241e GKH |
232 | static DEVICE_ATTR_RO(stable_pages_required); |
233 | ||
234 | static struct attribute *bdi_dev_attrs[] = { | |
235 | &dev_attr_read_ahead_kb.attr, | |
236 | &dev_attr_min_ratio.attr, | |
237 | &dev_attr_max_ratio.attr, | |
238 | &dev_attr_stable_pages_required.attr, | |
239 | NULL, | |
cf0ca9fe | 240 | }; |
d9e1241e | 241 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
242 | |
243 | static __init int bdi_class_init(void) | |
244 | { | |
245 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
246 | if (IS_ERR(bdi_class)) |
247 | return PTR_ERR(bdi_class); | |
248 | ||
d9e1241e | 249 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 250 | bdi_debug_init(); |
cf0ca9fe PZ |
251 | return 0; |
252 | } | |
76f1418b | 253 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 254 | |
26160158 JA |
255 | static int __init default_bdi_init(void) |
256 | { | |
257 | int err; | |
258 | ||
839a8e86 | 259 | bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE | |
b5c872dd | 260 | WQ_UNBOUND | WQ_SYSFS, 0); |
839a8e86 TH |
261 | if (!bdi_wq) |
262 | return -ENOMEM; | |
263 | ||
26160158 JA |
264 | err = bdi_init(&default_backing_dev_info); |
265 | if (!err) | |
266 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
976e48f8 | 267 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
268 | |
269 | return err; | |
270 | } | |
271 | subsys_initcall(default_bdi_init); | |
272 | ||
03ba3782 JA |
273 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
274 | { | |
275 | return wb_has_dirty_io(&bdi->wb); | |
276 | } | |
277 | ||
6467716a AB |
278 | /* |
279 | * This function is used when the first inode for this bdi is marked dirty. It | |
280 | * wakes-up the corresponding bdi thread which should then take care of the | |
281 | * periodic background write-out of dirty inodes. Since the write-out would | |
282 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
283 | * set up a timer which wakes the bdi thread up later. | |
284 | * | |
285 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
286 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
287 | * by delaying the wake-up. | |
6ca738d6 DB |
288 | * |
289 | * We have to be careful not to postpone flush work if it is scheduled for | |
290 | * earlier. Thus we use queue_delayed_work(). | |
6467716a AB |
291 | */ |
292 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) | |
293 | { | |
294 | unsigned long timeout; | |
295 | ||
296 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
5acda9d1 JK |
297 | spin_lock_bh(&bdi->wb_lock); |
298 | if (test_bit(BDI_registered, &bdi->state)) | |
299 | queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout); | |
300 | spin_unlock_bh(&bdi->wb_lock); | |
03ba3782 JA |
301 | } |
302 | ||
cfc4ba53 JA |
303 | /* |
304 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
305 | */ | |
306 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
307 | { | |
308 | spin_lock_bh(&bdi_lock); | |
309 | list_del_rcu(&bdi->bdi_list); | |
310 | spin_unlock_bh(&bdi_lock); | |
311 | ||
ef323088 | 312 | synchronize_rcu_expedited(); |
cfc4ba53 JA |
313 | } |
314 | ||
cf0ca9fe PZ |
315 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
316 | const char *fmt, ...) | |
317 | { | |
cf0ca9fe | 318 | va_list args; |
cf0ca9fe PZ |
319 | struct device *dev; |
320 | ||
69fc208b | 321 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
c284de61 | 322 | return 0; |
f1d0b063 | 323 | |
cf0ca9fe | 324 | va_start(args, fmt); |
19051c50 | 325 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 326 | va_end(args); |
c284de61 AB |
327 | if (IS_ERR(dev)) |
328 | return PTR_ERR(dev); | |
66f3b8e2 | 329 | |
cf0ca9fe | 330 | bdi->dev = dev; |
cf0ca9fe | 331 | |
03ba3782 | 332 | bdi_debug_register(bdi, dev_name(dev)); |
500b067c | 333 | set_bit(BDI_registered, &bdi->state); |
c284de61 AB |
334 | |
335 | spin_lock_bh(&bdi_lock); | |
336 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
337 | spin_unlock_bh(&bdi_lock); | |
338 | ||
455b2864 | 339 | trace_writeback_bdi_register(bdi); |
c284de61 | 340 | return 0; |
cf0ca9fe PZ |
341 | } |
342 | EXPORT_SYMBOL(bdi_register); | |
343 | ||
344 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
345 | { | |
346 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
347 | } | |
348 | EXPORT_SYMBOL(bdi_register_dev); | |
349 | ||
03ba3782 JA |
350 | /* |
351 | * Remove bdi from the global list and shutdown any threads we have running | |
352 | */ | |
353 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 354 | { |
03ba3782 JA |
355 | if (!bdi_cap_writeback_dirty(bdi)) |
356 | return; | |
357 | ||
358 | /* | |
fff5b85a | 359 | * Make sure nobody finds us on the bdi_list anymore |
03ba3782 | 360 | */ |
fff5b85a | 361 | bdi_remove_from_list(bdi); |
03ba3782 | 362 | |
5acda9d1 JK |
363 | /* Make sure nobody queues further work */ |
364 | spin_lock_bh(&bdi->wb_lock); | |
365 | clear_bit(BDI_registered, &bdi->state); | |
366 | spin_unlock_bh(&bdi->wb_lock); | |
367 | ||
03ba3782 | 368 | /* |
839a8e86 TH |
369 | * Drain work list and shutdown the delayed_work. At this point, |
370 | * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi | |
371 | * is dying and its work_list needs to be drained no matter what. | |
03ba3782 | 372 | */ |
839a8e86 TH |
373 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
374 | flush_delayed_work(&bdi->wb.dwork); | |
375 | WARN_ON(!list_empty(&bdi->work_list)); | |
c0ea1c22 | 376 | WARN_ON(delayed_work_pending(&bdi->wb.dwork)); |
66f3b8e2 JA |
377 | } |
378 | ||
592b09a4 JA |
379 | /* |
380 | * This bdi is going away now, make sure that no super_blocks point to it | |
381 | */ | |
382 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
383 | { | |
384 | struct super_block *sb; | |
385 | ||
386 | spin_lock(&sb_lock); | |
387 | list_for_each_entry(sb, &super_blocks, s_list) { | |
388 | if (sb->s_bdi == bdi) | |
95f28604 | 389 | sb->s_bdi = &default_backing_dev_info; |
592b09a4 JA |
390 | } |
391 | spin_unlock(&sb_lock); | |
392 | } | |
393 | ||
cf0ca9fe PZ |
394 | void bdi_unregister(struct backing_dev_info *bdi) |
395 | { | |
b6875734 | 396 | if (bdi->dev) { |
ccb6108f | 397 | bdi_set_min_ratio(bdi, 0); |
455b2864 | 398 | trace_writeback_bdi_unregister(bdi); |
8c4db335 JA |
399 | bdi_prune_sb(bdi); |
400 | ||
839a8e86 | 401 | bdi_wb_shutdown(bdi); |
76f1418b | 402 | bdi_debug_unregister(bdi); |
b6875734 | 403 | device_unregister(bdi->dev); |
cf0ca9fe PZ |
404 | bdi->dev = NULL; |
405 | } | |
406 | } | |
407 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 408 | |
6467716a AB |
409 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
410 | { | |
411 | memset(wb, 0, sizeof(*wb)); | |
412 | ||
413 | wb->bdi = bdi; | |
414 | wb->last_old_flush = jiffies; | |
415 | INIT_LIST_HEAD(&wb->b_dirty); | |
416 | INIT_LIST_HEAD(&wb->b_io); | |
417 | INIT_LIST_HEAD(&wb->b_more_io); | |
f758eeab | 418 | spin_lock_init(&wb->list_lock); |
839a8e86 | 419 | INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn); |
6467716a AB |
420 | } |
421 | ||
e98be2d5 WF |
422 | /* |
423 | * Initial write bandwidth: 100 MB/s | |
424 | */ | |
425 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) | |
426 | ||
b2e8fb6e PZ |
427 | int bdi_init(struct backing_dev_info *bdi) |
428 | { | |
03ba3782 | 429 | int i, err; |
b2e8fb6e | 430 | |
cf0ca9fe PZ |
431 | bdi->dev = NULL; |
432 | ||
189d3c4a | 433 | bdi->min_ratio = 0; |
a42dde04 | 434 | bdi->max_ratio = 100; |
eb608e3a | 435 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
03ba3782 | 436 | spin_lock_init(&bdi->wb_lock); |
66f3b8e2 | 437 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
438 | INIT_LIST_HEAD(&bdi->work_list); |
439 | ||
440 | bdi_wb_init(&bdi->wb, bdi); | |
441 | ||
b2e8fb6e | 442 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
908c7f19 | 443 | err = percpu_counter_init(&bdi->bdi_stat[i], 0, GFP_KERNEL); |
04fbfdc1 PZ |
444 | if (err) |
445 | goto err; | |
446 | } | |
447 | ||
448 | bdi->dirty_exceeded = 0; | |
e98be2d5 WF |
449 | |
450 | bdi->bw_time_stamp = jiffies; | |
451 | bdi->written_stamp = 0; | |
452 | ||
7381131c | 453 | bdi->balanced_dirty_ratelimit = INIT_BW; |
be3ffa27 | 454 | bdi->dirty_ratelimit = INIT_BW; |
e98be2d5 WF |
455 | bdi->write_bandwidth = INIT_BW; |
456 | bdi->avg_write_bandwidth = INIT_BW; | |
457 | ||
20ae0079 | 458 | err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL); |
04fbfdc1 PZ |
459 | |
460 | if (err) { | |
461 | err: | |
4b01a0b1 | 462 | while (i--) |
04fbfdc1 | 463 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
464 | } |
465 | ||
466 | return err; | |
467 | } | |
468 | EXPORT_SYMBOL(bdi_init); | |
469 | ||
470 | void bdi_destroy(struct backing_dev_info *bdi) | |
471 | { | |
472 | int i; | |
473 | ||
ce5f8e77 | 474 | /* |
1a1e4530 TH |
475 | * Splice our entries to the default_backing_dev_info. This |
476 | * condition shouldn't happen. @wb must be empty at this point and | |
477 | * dirty inodes on it might cause other issues. This workaround is | |
478 | * added by ce5f8e779519 ("writeback: splice dirty inode entries to | |
479 | * default bdi on bdi_destroy()") without root-causing the issue. | |
480 | * | |
481 | * http://lkml.kernel.org/g/1253038617-30204-11-git-send-email-jens.axboe@oracle.com | |
482 | * http://thread.gmane.org/gmane.linux.file-systems/35341/focus=35350 | |
483 | * | |
484 | * We should probably add WARN_ON() to find out whether it still | |
485 | * happens and track it down if so. | |
ce5f8e77 JA |
486 | */ |
487 | if (bdi_has_dirty_io(bdi)) { | |
488 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
489 | ||
f758eeab | 490 | bdi_lock_two(&bdi->wb, dst); |
ce5f8e77 JA |
491 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); |
492 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
493 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
f758eeab CH |
494 | spin_unlock(&bdi->wb.list_lock); |
495 | spin_unlock(&dst->list_lock); | |
ce5f8e77 | 496 | } |
66f3b8e2 | 497 | |
cf0ca9fe PZ |
498 | bdi_unregister(bdi); |
499 | ||
c0ea1c22 | 500 | WARN_ON(delayed_work_pending(&bdi->wb.dwork)); |
7a401a97 | 501 | |
b2e8fb6e PZ |
502 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
503 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 | 504 | |
eb608e3a | 505 | fprop_local_destroy_percpu(&bdi->completions); |
b2e8fb6e PZ |
506 | } |
507 | EXPORT_SYMBOL(bdi_destroy); | |
508 | ||
c3c53206 JA |
509 | /* |
510 | * For use from filesystems to quickly init and register a bdi associated | |
511 | * with dirty writeback | |
512 | */ | |
b4caecd4 | 513 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name) |
c3c53206 | 514 | { |
c3c53206 JA |
515 | int err; |
516 | ||
517 | bdi->name = name; | |
b4caecd4 | 518 | bdi->capabilities = 0; |
c3c53206 JA |
519 | err = bdi_init(bdi); |
520 | if (err) | |
521 | return err; | |
522 | ||
02aa2a37 KC |
523 | err = bdi_register(bdi, NULL, "%.28s-%ld", name, |
524 | atomic_long_inc_return(&bdi_seq)); | |
c3c53206 JA |
525 | if (err) { |
526 | bdi_destroy(bdi); | |
527 | return err; | |
528 | } | |
529 | ||
530 | return 0; | |
531 | } | |
532 | EXPORT_SYMBOL(bdi_setup_and_register); | |
533 | ||
3fcfab16 AM |
534 | static wait_queue_head_t congestion_wqh[2] = { |
535 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
536 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
537 | }; | |
0e093d99 | 538 | static atomic_t nr_bdi_congested[2]; |
3fcfab16 | 539 | |
1faa16d2 | 540 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
541 | { |
542 | enum bdi_state bit; | |
1faa16d2 | 543 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 544 | |
1faa16d2 | 545 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
546 | if (test_and_clear_bit(bit, &bdi->state)) |
547 | atomic_dec(&nr_bdi_congested[sync]); | |
4e857c58 | 548 | smp_mb__after_atomic(); |
3fcfab16 AM |
549 | if (waitqueue_active(wqh)) |
550 | wake_up(wqh); | |
551 | } | |
552 | EXPORT_SYMBOL(clear_bdi_congested); | |
553 | ||
1faa16d2 | 554 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
555 | { |
556 | enum bdi_state bit; | |
557 | ||
1faa16d2 | 558 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
559 | if (!test_and_set_bit(bit, &bdi->state)) |
560 | atomic_inc(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
561 | } |
562 | EXPORT_SYMBOL(set_bdi_congested); | |
563 | ||
564 | /** | |
565 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 566 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
567 | * @timeout: timeout in jiffies |
568 | * | |
569 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
570 | * write congestion. If no backing_devs are congested then just wait for the | |
571 | * next write to be completed. | |
572 | */ | |
8aa7e847 | 573 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
574 | { |
575 | long ret; | |
52bb9198 | 576 | unsigned long start = jiffies; |
3fcfab16 | 577 | DEFINE_WAIT(wait); |
8aa7e847 | 578 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
579 | |
580 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
581 | ret = io_schedule_timeout(timeout); | |
582 | finish_wait(wqh, &wait); | |
52bb9198 MG |
583 | |
584 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
585 | jiffies_to_usecs(jiffies - start)); | |
586 | ||
3fcfab16 AM |
587 | return ret; |
588 | } | |
589 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 590 | |
0e093d99 MG |
591 | /** |
592 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes | |
593 | * @zone: A zone to check if it is heavily congested | |
594 | * @sync: SYNC or ASYNC IO | |
595 | * @timeout: timeout in jiffies | |
596 | * | |
597 | * In the event of a congested backing_dev (any backing_dev) and the given | |
598 | * @zone has experienced recent congestion, this waits for up to @timeout | |
599 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
600 | * or a write to complete. | |
601 | * | |
25985edc | 602 | * In the absence of zone congestion, cond_resched() is called to yield |
0e093d99 MG |
603 | * the processor if necessary but otherwise does not sleep. |
604 | * | |
605 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
606 | * it is the number of jiffies that were still remaining when the function | |
607 | * returned. return_value == timeout implies the function did not sleep. | |
608 | */ | |
609 | long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
610 | { | |
611 | long ret; | |
612 | unsigned long start = jiffies; | |
613 | DEFINE_WAIT(wait); | |
614 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
615 | ||
616 | /* | |
617 | * If there is no congestion, or heavy congestion is not being | |
618 | * encountered in the current zone, yield if necessary instead | |
619 | * of sleeping on the congestion queue | |
620 | */ | |
621 | if (atomic_read(&nr_bdi_congested[sync]) == 0 || | |
57054651 | 622 | !test_bit(ZONE_CONGESTED, &zone->flags)) { |
0e093d99 MG |
623 | cond_resched(); |
624 | ||
625 | /* In case we scheduled, work out time remaining */ | |
626 | ret = timeout - (jiffies - start); | |
627 | if (ret < 0) | |
628 | ret = 0; | |
629 | ||
630 | goto out; | |
631 | } | |
632 | ||
633 | /* Sleep until uncongested or a write happens */ | |
634 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
635 | ret = io_schedule_timeout(timeout); | |
636 | finish_wait(wqh, &wait); | |
637 | ||
638 | out: | |
639 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
640 | jiffies_to_usecs(jiffies - start)); | |
641 | ||
642 | return ret; | |
643 | } | |
644 | EXPORT_SYMBOL(wait_iff_congested); | |
3965c9ae WL |
645 | |
646 | int pdflush_proc_obsolete(struct ctl_table *table, int write, | |
647 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
648 | { | |
649 | char kbuf[] = "0\n"; | |
650 | ||
4c3bffc2 | 651 | if (*ppos || *lenp < sizeof(kbuf)) { |
3965c9ae WL |
652 | *lenp = 0; |
653 | return 0; | |
654 | } | |
655 | ||
656 | if (copy_to_user(buffer, kbuf, sizeof(kbuf))) | |
657 | return -EFAULT; | |
658 | printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n", | |
659 | table->procname); | |
660 | ||
661 | *lenp = 2; | |
662 | *ppos += *lenp; | |
663 | return 2; | |
664 | } |